Federated Deep Reinforcement Learning for Energy-Adaptive Smart Manufacturing: A Comparative Multi-Factory Analysis of Edge–Cloud Industrial IoT Architectures

Abstract

The rapid convergence of artificial intelligence, Industrial Internet of Things (IIoT), and sustainable manufacturing has transformed industrial production systems into highly interconnected cyber-physical ecosystems. However, despite extensive deployment of smart manufacturing infrastructures, industrial facilities continue to experience substantial inefficiencies associated with energy consumption, predictive maintenance, adaptive scheduling, and real-time process optimization. Existing industrial automation frameworks frequently rely on centralized machine learning architectures that suffer from scalability constraints, data privacy concerns, latency accumulation, and limited adaptability across heterogeneous production environments. This study develops and comparatively evaluates a federated deep reinforcement learning framework for energy-adaptive smart manufacturing across two distinct industrial IoT architectures: centralized cloud-based manufacturing intelligence and distributed edge–cloud collaborative intelligence. The research integrates computational simulation, industrial benchmark datasets, predictive maintenance streams, and adaptive scheduling environments to examine how decentralized learning mechanisms influence operational efficiency, energy optimization, and manufacturing resilience. Comparative computational evaluation demonstrates that the edge–cloud federated architecture significantly reduces decision latency, improves predictive maintenance responsiveness, enhances energy efficiency, and increases production adaptability under dynamic operational conditions. The findings further indicate that federated reinforcement learning improves cross-factory knowledge transfer while preserving industrial data sovereignty and cybersecurity integrity. This article contributes to engineering scholarship by integrating systems engineering theory, industrial AI optimization, distributed computation, and sustainable manufacturing analysis into a unified technological framework. The study also provides operational implications for Industry 4.0 implementation, intelligent manufacturing scalability, and environmentally sustainable industrial transformation.